习题内外表示异质融合的知识追踪模型

现有知识追踪研究大多使用习题蕴涵的知识点等内隐信息或历史交互数据等外显信息建模习题表示，没有注意到内外信息的异质性特征，缺乏对习题内外信息的异质融合。针对上述问题，提出了融合内外异质信息的知识追踪模型。首先，基于知识点等内隐信息，计算历史知识点与当前知识点之间的相关程度，刻画历史知识点对当前知识点的影响，建模习题的内隐表示；其次，基于交互数据等外显信息，计算历史习题与当前习题之间的相关程度，获取历史习题对当前习题的影响，建模习题的外显表示；再次，基于上述习题的内外表示，使用通道注意力机制融合得到习题的内外异质表示，从而预测学习者的作答表现。为了验证提出模型的性能和有效性，选取了四个相关的基线模型，在三个真实数据集上进行了对比实验。实验结果表明：在性能方面，提出的模型在多个评价指标上均取得较好的效果；在有效性方面，消融实验证明了提出的模型可以更好地根据内外信息建模习题表示；在应用方面，设计智慧学习环境证明了提出的模型在实际教学场景中的可用性。

Knowledge tracing via heterogeneous fusion of exercises’ internal and external representations

Most of the existing knowledge tracing research use implicit information such as concepts contained in exercises or explicit information such as historical interaction data to model exercises, they don''t pay attention to the heterogeneity of internal and external information, and lack heterogeneous fusion of internal and external information. To address these issues, this paper proposed a knowledge tracing model that integrated heterogeneous internal and external information. Firstly, this model calculated the relevance between historical concepts and current concepts based on implicit information like concepts, depicted the influence of historical concepts on the current ones and modeled the exercises'' implicit representation. Secondly, it computed the relevance between historical exercises and current exercises by explicit information like interaction data, captured the impact of historical exercises on the current ones and established the exercises'' explicit representation. Furthermore, this paper utilized the channel attention mechanism on the aforementioned internal and external exercises representation, and achieved a fusion of heterogeneous information to create the exercises'' heterogeneous representation, enabling the prediction of learners'' performance. To validate the performance and effectiveness of the proposed model, this paper conducted comparative experiments on three real-world datasets using four relevant baseline models. The experimental results demonstrate that the proposed model achieves superior performance on multiple evaluation metrics. Additionally, ablation experiments confirm the effectiveness of the proposed model in better modeling exercise representations by incorporating both internal and external information. In the terms of applications, a smart learning environment is designed to prove the advantages of the proposed model in actual teaching scenarios.